AI Agents Are Changing Our Relationship With Data
I would argue that at the heart of Information Technology lies our evolving relationship with information.
The first wave brought unprecedented access to data, democratising access to knowledge.
Now, we’re entering an era of highly personalised, contextualised data tailored to each user — an audience of one.
We’ve moved beyond simple search, now synthesising data from diverse sources via systems that understand, process, and act on information in ways that mimic human cognition.
Old Search delivered URLs leaving the user with the task of curating URLs, sifting out sponsored content and poor content with good SEO. Search 2.0 has a wider search, curates and synthesis various sources of data based on the data format defined by the user.
AI Agents & How Data is Surfaced
So, AI Agents have become the new vehicle through which search and recommendation systems are delivered, offering a seamless integration of perception, control, decision-making, and action.
As illustrated in recent frameworks, AI Agents operate through core modules that enable their functionality.
Recommendations and search rankings have evolved with mobile search and commerce, driven by technology and user needs.
Considering the image above, it shows the role of an AI Agent within a recommender system (RS), highlighting its central position in managing interactions and optimising recommendations.
It depicts four key components —
- User Interaction,
- Representation Optimisation,
- Simulation Environment, and
- System Integration
User Interaction emphasises the AI Agent’s role as an intermediary to enhance communication between users and the RS, improving interactivity.
Representation Optimisation focuses on the AI Agent’s ability to refine user and item representations, enhancing the system’s understanding and personalisation.
Considering the image below, it emphasises the AI Agent’s central interaction with various components.
It features a Search System which can underpin an AI Agent, including Match, Rank, and Re-Rank processes.
Five key functions are:
- Task Decomposer,
- User Simulator,
- Query Rewriter,
- Results Synthesis, and
- Action Executor.
Task Decomposer breaks down complex search tasks into subtasks to enhance overall efficiency and accuracy.
User Simulator acts as a proxy to simulate user interactions and provide feedback for system improvement or evaluation.
Query Rewriter refines user queries to ensure clearer and more precise search results, while Results Synthesis summarises relevant text to assist users in decision-making.
Action Executor enables the agent to interact with tools and application interfaces, gathering necessary information to optimise search outcomes.
Planning
We all know by now that AI Agents make use of one or more Large Language Models (LLMs) as it’s backbone and this combined with using an LLM with an agentic framework have demonstrated impressive ability in complex task decomposition and planning; the bedrock of autonomy.
The planning component of AI Agents has been extracted and used in isolation; giving rise to the notion of Agentic Workflows with human supervision. And the human acts as a gatekeeper to edit, accept or reject the sub-steps which constitutes the AI Agent planning.
This graphic below gives a clear and organised overview of research on LLM-based agent planning, focusing on recent efforts to enhance planning skills.
It groups existing studies into five categories:
- Task Decomposition,
- Plan Selection,
- External Module,
- Reflection, and
- Memory
AI Agents are seen as smart systems that can handle specific tasks by observing their surroundings, planning, and taking action. Planning is a key ability for AI Agents, requiring a deep understanding, reasoning and decision-making skills.
Chief Evangelist @ Kore.ai | I’m passionate about exploring the intersection of AI and language. From Language Models, AI Agents to Agentic Applications, Development Frameworks & Data-Centric Productivity Tools, I share insights and ideas on how these technologies are shaping the future.
